Reassessing CTV-to-PTV Margins by Analysing the Patient Motion of Patients Treated with Intracranial SRS

Jeremy Hughes (00:03):

So today I’ll be talking about CTV-to-PTV margins – very exciting. I know by basically looking at patient motion during treatment, I originally gave this talk for a more physics audience, so I do apologize in advance. I’ll cover a lot of the fun equations. How about that? We’ve got an institution agreement with Vision RT. I’m related to this. So let’s start with the basics. I’m sure you are well aware. We have a GTV that we want to hit. We put a small margin on that. That’s our CTV which is taking into account a lot of the microscopic tumor cells that we want to hit. We then do another expansion to the PTV. This takes into account a lot of the uncertainties, you know, machine geometry uncertainties, treatment planning uncertainties, just so that we’re confident that we are delivering the dose to the CTV.

Jeremy Hughes (01:03):

In intercranial SRS cases, that’s right in the middle of your brain, you want to be sure that you are delivering the dose to the tumor, but you don’t want to be delivering too much because you don’t want to be delivering dose to healthy brain tissue. So yeah, so in these cases, you want your CTV-to-PTV margin to be basically as tight as you are comfortable with. So when I made this talk our SRS that we deliver on our TrueBeam Linux with onboard imaging we used a margin of one and a half mils. So the question is, can we safely reduce this number?

Jeremy Hughes (02:03):

This is just a brief overview, we’ve got true beams. We’ve on board imaging typically three treatment arcs at catch 0, 45, 315, half a on the couch kicks. We do pre-beam imaging, a cone beam CT at couch zero, and we are still verifying with MVKV imaging at those couch kicks. We use AlignRT to monitor these patients with an open face mask with a one mil toll and a 0.7 degree toll.

Jeremy Hughes (02:34):

So I basically looked at a lot of our SRS patient data to see how they move. You’ve got all the real-time deltas being saved on the backend that you can pull out and all analyze. So I wrote a little script that does that. I also briefly looked at camera occlusion very briefly. It’s more of a sidebar than anything here.

Jeremy Hughes (02:58):

So this is a fun little equation. There’s lots of different ways to assess A CTV-to-PTV margin. This is just one of them. The main takeaway from this is that you’ve got systematic error and random error, and patient motion really falls into that random error component. There’s just some examples there for you. It should be noted that there is this equation that will get you a number, but in all of like, you know, the documentation and all the reports and everything they’re saying, you know, don’t take this number and just run with it. You know, this is in consultation with all the ROIs. There’s a lot of different factors in play here. However, the formula does not define an uncontestable correct margin to use clinically. And throughout this report, they restated that at least 10 times different ways.

Jeremy Hughes (03:58):

So we’ve got AlignRT out at Peter Mac, I’m in the Moorabbin campus, but we’ve got them all over Peter Mac. And it’s very basic. So there’s our little open face mask. We’ve got a ROI that we place just on that patient opening. You capture a reference image and then it basically monitors how much your live ROI is different to that reference image that you’ve captured. AlignRT is taking a lot of information in the backend. It does this every maybe 0.1 of a second or so. You’ve got all your delta patient movements, you’ve got percentage overlap there which basically is what percentage of your ROI is AlignRT using to do its calculation. So sometimes with camera blocking or something that will drop down. But you’ll still get the, the data.

Jeremy Hughes (05:02):

If you want, you can make some pretty graphs, and I love pretty graphs. This is basically a whole-patient treatment. There’s a lot of information. I don’t know why it’s automatically moving to the next one, but that’s fine. So you’ve got the whole patient treatment. There’s your three arcs – Orange is beam on, so there’s a first arc being delivered, second arc being delivered, third arc being delivered. And then you’ve also got your AlignRT thresholds here. And then, like reference images being taken for AlignRT. And you can see how the patient sort of moves in reference to that reference image over the course of the treatment. So zooming in on one arc delivery, you can see in this case, you’ve got vertical longitudinal lateral components, and then there’s your rotation pitch and roll. There’s your magnitude error as well. And I’ve just pulled out the couch rotation. So this is for couch zero and your little percentage overlap, which I’ve sort of used as a surrogate for camera occlusion. There’s a lot more to do in this, but basically, you can say, Hey, the ROI, there’s less ROI being used here in this little region. And you can kind of see it corresponds with a jump in the parameters of the patient motion.

Jeremy Hughes (06:29):

So there’s a lot of math’s that you can go into which I want. But basically I looked at our patient cohort. I did a lot of medians. I did a lot of standard deviations of how much they’re moving inside the actual beam itself. And you can kind of come with a systematic error of your patient cohort being around, Hey, let’s 0.1 degree and then your standard deviation, that’s your spread as well. So that’s only analyzing when the beam is turned on. And so it’s ignoring all those times where the patient has triggered an Interlochen AlignRT.

Jeremy Hughes (07:16):

I basically also looked at the percentage overlap as well. So I put thresholds to the data. So going back here, I looked at it and I basically said, all right, let’s analyze all of the data. Let’s analyze only when the data is say, above 70%. So that would ignore that data as well, just to see how the numbers changed basically. And all in all, it’s sitting at around, you know, 0.1, roughly point 0.1 mil a little bit more in the long, which is you, I feel like you sort of expect just from looking at a delivery you can see the long is kicking off a bit more.

Jeremy Hughes (08:04):

In these margin calculations, of course there’s a lot of caveats as there always is. In this one, I’m purely looking at the ISO being in the middle of a target. I’m not looking at lots of different targets with one iso. So, not getting into some of that funky rotational stuff that Michael was talking about in the previous talk. And there’s a whole lot of numbers, as he was saying, linear algebra, add them up in a quad chart, do some analysis and that helps you sort of come to a margin at the end of it. So this is looking at, I guess, more of a geometric margin. So really looking at that the geometric misses in theory, again, like I said at the beginning, there’s a lot that goes into the actual conversation about what a CTV-to-PTV margin is. But that definitely, I guess, put our mind at ease, you know, in talking to the ROS who are very keen on reducing that margin down to be like, well, one mil isn’t so far out from our calculated margin with all of the tolerances and assumptions inbuilt into it. So all of that to say we looked at a lot of our patient cohort and did our math’s as you saw before. And it really helped us to, I guess, be like, well, one mill for these cases is well acceptable, looking at our data. And camera occlusion didn’t really have a lot of impacts in the, I guess, patient information as a whole. But I think that’s another scope that you, we want to investigate further. I will say, just because I forgot to mention it before, all the real-time delta analysis is quite nice as well. You do have to extract it out from the machine, which I found was the, I guess, the biggest barrier for entry for just my analysis, just because you do have to like physically go down there and like click buttons. I tried to automate it, but because of that, I guess we’ve treated a lot of patients here, but I only got to analyze, you know, a subset of them basically. But yeah ongoing, this could be an ongoing thing. You could in theory, extend this out to all your patients treated with surface guided to, you know, really retrospectively just analyse the patient motion. But it is the case that you still need to physically go and export it, which I think is the main limiting factor of this. Yeah. Thank you.